strong. Karl et al. also used station data to extend their analysis to the multidecade-to-century time scale. Their findings included a 4 to 5 percent increase of both solid and total precipitation rates over northern Canada during the past four decades, century-scale increases of precipitation over southern Canada and the contiguous United States, and a decrease in the proportion of precipitation falling as snow in southern Canada. These findings are generally consistent with the changes that have been hypothesized to accompany a greenhouse warming. Karl et al. also found that the unprecedented warmth of the 1980s in Alaska was accompanied by a 10 percent increase of annual precipitation in that region. Leathers and Robinson (1993) have examined the winter region (the central United States) in more detail, finding that the concurrent 500 mb height anomalies are generally collocated with the snow/temperature anomalies during December but not during January and February.
The general coincidence of the "marginal snow zone" and the areas of strongest warming over the past several decades (Figure 4) deserves further comment with regard to the possible nature of the forcing. The "land-leading-ocean" feature is characteristic of large-scale forcing such as global warming or the response to major volcanic events. By contrast, natural low-frequency variations will generally manifest themselves in an "ocean-leading-land" pattern because the ocean is the low-frequency source of the forcing on the near-surface atmospheric temperature. The extent to which a large-scale forced temperature response is amplified by the retreat of snow over land is one of the key unknowns in the interpretation of a pattern such as that in Figure 4.
Decadal-scale summaries of snowfall in China have been compiled by Li (1987), who found a decrease of snowfall over China during the 1950s followed by an increase during the 1960s and 1970s; the late 1970s had the largest running-mean values of the 30-year period of record (Figure 5). Li noted a general correspondence between the time series of global mean temperature and mean snow depth in China (as well as an apparent association between years of heavy
snow and ENSO events). Similar compilations have been made for other regions such as the Swiss Alps (Lang and Rohrer, 1987), although the representativeness of such time series is a key issue. From the standpoint of areal coverage, the snowfall records that are potentially most valuable are those of the former Soviet Union. The status of these data is uncertain.
The longest time series of snow cover are generally those coming from single stations (e.g., the Tokyo record dating back to the 1630s (Lamb, 1977)) or the records of annual precipitation derived from polar ice cores (e.g., the Greenland Ice Sheet Project's work in Antarctica and Greenland, as in Alley et al., 1993). The cores represent potentially valuable records of regional snowfall if the spatial representativeness of the point data can be established. The South Pole data, for example, have been used to deduce an increase of snowfall from the 1700s to the early twentieth century (Giovinetto and Schwertfeger, 1966; Lamb, 1977). Analysis of GISP's Greenland cores has only recently begun (Mayewski et al., 1993).
Another type of data for which analyses are in the early phases is the runoff (stream-flow) data for the high-latitude rivers that are fed primarily by snow melt. Mysak et al. (1990) have proposed a mechanism by which an interdecadal (approximately 20-year) cycle results from a feedback loop involving high-latitude precipitation, runoff, arctic sea-ice export to the North Atlantic, and ocean salinity/temperature anomalies. The linkages involving snowfall and runoff have yet to be thoroughly evaluated. In view of the potential implications of this cycle for the global ocean circulation, quantitative analyses of high-latitude snowfall and runoff are being assigned high priority in the upcoming ACSYS (Arctic Climate System) component of the World Climate Research Program. The extension of such analyses to include soil moisture and land surface temperatures also appears to merit high priority in the context of possible greenhouse-induced changes over decadal time scales.
Finally, glaciers are ultimately attributable to continental snowfall over time scales of decades to millennia. The advance and retreat of glaciers have long been regarded as proxy indicators of climate change over these time scales. In the context of global change, the key properties of glaciers are extent, ice volume, and mass balance. Temporal changes in these glacial properties can be complex functions of temperature, precipitation, the seasonality of temperature and precipitation, and topography. Nevertheless, the advance and retreat of glaciers are known to be consistent with century-scale changes of regional climate, e.g., the Little Ice Age.
Field measurements of glaciers are now assembled and reported by the World Glacier Monitoring Service in Zürich. These data, which are summarized by Haberli et al. (1989), include direct measurements of the mass balance of approximately 75 glaciers in the Northern Hemisphere. Statistics on